Sheetfed scanners are widely used for scanning stacks of loose pages at high speed. The scanhead in the sheet-fed scanners is stationary and the page is fed with an automatic document feeder. When dust particles get stuck onto the scanner glass, they reflect the incident light and cause vertical streaks in the scanned images. These artifacts are known as dust streaks. We have developed a method for detecting dust streaks and the results were published in our previous paper [1]. In this study, we have refined our features for dust detection and added features for detecting tables. In addition, we looked into two methods for healing defective images: an exemplar based method and a diffusion based method. We applied these methods to remove dust streaks, punch holes, and torn corners from scanned images.
Dust in the scanner may cause vertical streaks in the scanned image since it reflects some part of the incident light. In this paper, we propose a method for detecting streaks in the scanned images that are a direct result of dust on the scanner glass. This lets customers resolve the issue without calling the maintenance. The solution includes denoising, conversion to opponent color space, calculation of ΔE', calculation of features, and classification. We denoise the image in order to remove halftones in case the image was halftoned. Opponent color space lets us look separately at luminance channel and chrominance channels. We have developed three features that use the data in luminance and chrominance channels. Eventually, we will use these features to detect streaks, and distinguish them from content.